This paper describes further enhancement of the Normal Ray Refinement (NRR) technique and its implementation in the adaptive Cartesian framework of the Unified Flow Solver (UFS). Brief introduction to the UFS Navier-Stokes (NS) solver with an Immersed Boundary Method (IBM) is given. Demonstrations are provided of NRR grid generation algorithms based on normal rays with constant and variable refinement on 2D and 3D shapes. Low-and high-order accuracy inter-ray communications methods are implemented in 2Dand results for different shapes are demonstrated and compared to solutions obtained with uniform boundary grids. Capabilities for adaptive normal ray placement are developed in 2D and 3D thus providing ways for NRR automation when fully implemented. Results of flow solution for 2D geometries with statically placed and adaptive normal ray placement show good agreement with uniform boundary layer results. Implementation of the basic framework for NRR over 3D shapes is described and examples are shown. In prior study, the NRR technique was shown to provide accurate prediction of incompressible, attached viscous flows. In the current study the NRR methodology, implemented in NASCART-GT solver, is shown to provide accurate prediction in supersonic flow and low-speed, separated flow around a circular cylinder. It is found that the NRR technique is robustly implementable in different frameworks and it offers significant advantages in terms of greatly reduced number of cells in boundary layers when using Cartesian grids.